Egocentric-10K: 10,000 Hours of Industrial First-Person Video

Egocentric-10K contains 10,000 hours of first-person video footage comprising 1.08 billion frames recorded exclusively in real-world factory environments. Developed by builddotai and updated in February 2026, it provides a massive-scale resource for industrial action recognition and hand-tracking research.

Use Cases

• Hand-tracking model development using high-visibility hand segments
• Industrial action recognition for manufacturing and assembly tasks
• Egocentric video summarization within factory workflows

Strengths

• 1.08 billion total frames
• 10,000 hours of video duration
• Exclusively recorded in real factory environments
• High density of active hand manipulation frames

Limitations

• Domain-specific focus on factories may limit generalizability to domestic or outdoor egocentric tasks
• Massive scale (1.08 billion frames) presents significant storage and computational overhead for users

Provenance

Source

builddotai

Collection Method

Sensor-based recording using egocentric cameras in real-world industrial facilities.

Freshness

Last updated February 2026.

Geography

United States